Corpora of Non-Linguistic Symbol Systems

Humans have been writing for over 5,000 years, but in addition to linguistic symbol systems there have been many non-linguistic systems. Some examples include mathematical symbology, European heraldry, barn stars, Mesopotamian deity symbols and totem poles. While writing represents natural language units such as phonemes, syllables, morphemes or in some cases words, non-linguistic systems represent other, non-linguistic, information. Thus, mathematical symbols represent mathematical operations, functions, variables and the like. Note that it does not matter that one can read a mathematical equation using words; the elements of the equation do not represent words, or any other linguistic elements. Within the past few years, two high-profile papers have claimed to provide statistical methods to distinguish writing from non-linguistic symbol systems. The first, (Rao et al., 2009), used bigram conditional entropy to argue that the symbols used by the Indus Valley civilization constituted a writing system. The second, (Lee et al., 2010), used a different technique also based on conditional entropy to argue that Pictish symbols, found on a few hundred standing stones in Scotland, were part of a heretofore unrecognized writing system. Both of these papers were very favorably reported in the popular science press. The problem is that the techniques reported in the cited papers do not provide evidence that a system is linguistic: for example, they are easily fooled by artificial systems that are generated by non-uniform memoryless random processes. But the deeper and more important point is that in order to test any statistical method that purports to distinguish writing from non-writing, one surely needs a set of corpora of clear non-linguistic symbol systems. Few such corpora exist. The project reported in this paper fills that void by developing electronic corpora of known non-linguistic systems. To date we have developed corpora of the following systems: European heraldry; totem poles; Mesopotamian deity symbols (kudurrus) (Seidl, 1989); Vinča symbols (Winn, 1981); Pictish symbols; mathematical equations downloaded from arXiv.org; weather icon sequences from 5-day forecasts downloaded from wunderground.com, and Pennsylvania German barn stars (also known as “hex signs”) (Graves, 1984). Corpus sizes range from several hundred to several tens of thousands of symbols. All corpora are encoded using an XML-markup scheme based in part on the Text Encoding Initiative (tei-c.org) conventions. The corpora will be released under an open-source license via the Linguistic Data Consortium.